Lecture 14 - NESTED ANOVA

Bill Perry

Introduction

This analysis examines the effects of varying sea urchin densities on the percentage cover of filamentous algae. The experiment was designed with four urchin density treatments (control, 66% of original density, 33% of original density, and all urchins removed) nested within four random patches. Five replicate quadrats were measured within each treatment-patch combination.

The traditional nested ANOVA approach can be implemented using a linear mixed-effects model, which provides a more flexible framework for analyzing hierarchical designs. In this case, we’ll use the lme4 package to fit a model where treatment is a fixed effect and patch is a random effect nested within treatment.

Data Overview

The dataframe contains 80 observations with the following variables:

  • TREAT: Urchin density treatment (Control, 66% Density, 33% Density, Removed)
  • PATCH: Random patches (1-16) where treatments were applied
  • QUAD: Replicate quadrats within each treatment-patch combination
  • ALGAE: Percentage cover of filamentous algae (response variable)
# Create a summary table with flextable

summary_stats
# A tibble: 4 × 7
  TREAT           n  mean    sd    se   min   max
  <fct>       <int> <dbl> <dbl> <dbl> <dbl> <dbl>
1 Control        20   1.3  3.18 0.711     0    13
2 66% Density    20  21.6 25.1  5.62      0    79
3 33% Density    20  19   25.7  5.74      0    71
4 Removed        20  39.2 28.7  6.41      0    83
# 
# summary_stats %>%
#   select(TREAT, n, mean, sd, se, min, max) %>%
#   flextable() %>%
#   set_header_labels(
#     TREAT = "Treatment",
#     n = "N",
#     mean = "Mean",
#     sd = "SD",
#     se = "SE",
#     min = "Min",
#     max = "Max"
#   ) %>%
#   colformat_double(j = c("mean", "sd", "se", "min", "max"), digits = 2) %>%
#   autofit() %>%
#   add_header_lines("Summary statistics of algae cover (%) across treatments") %>%
#   theme_box()

Mixed Model Analysis

In this experimental design, PATCH is nested within TREAT because each patch received only one treatment level. This hierarchical design is well-suited for analysis using linear mixed-effects models.

Model Specification

We’ll use the following model specification:

\(ALGAE_{ijk} = \mu + \alpha_i + \beta_{j(i)} + \epsilon_{ijk}\)

Where: - \(\mu\) is the overall mean - \(\alpha_i\) is the fixed effect of treatment \(i\) - \(\beta_{j(i)}\) is the random effect of patch \(j\) nested within treatment \(i\) - \(\epsilon_{ijk}\) is the residual error for quadrat \(k\) in patch \(j\) within treatment \(i\)

In lme4, this model is specified as

# Fit the mixed model
mixed_model <- lmer(ALGAE ~ TREAT + (1|TREAT:PATCH), data = andrew)

# Display model summary
summary(mixed_model)
Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: ALGAE ~ TREAT + (1 | TREAT:PATCH)
   Data: andrew

REML criterion at convergence: 682.2

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-1.9808 -0.3106 -0.1093  0.2831  2.5910 

Random effects:
 Groups      Name        Variance Std.Dev.
 TREAT:PATCH (Intercept) 294.3    17.16   
 Residual                298.6    17.28   
Number of obs: 80, groups:  TREAT:PATCH, 16

Fixed effects:
                 Estimate Std. Error     df t value Pr(>|t|)  
(Intercept)         1.300      9.408 12.000   0.138   0.8924  
TREAT66% Density   20.250     13.305 12.000   1.522   0.1539  
TREAT33% Density   17.700     13.305 12.000   1.330   0.2081  
TREATRemoved       37.900     13.305 12.000   2.849   0.0147 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) TREAT6D TREAT3D
TREAT66%Dns -0.707                
TREAT33%Dns -0.707  0.500         
TREATRemovd -0.707  0.500   0.500

ANOVA Table

The ANOVA table for the mixed model:

# Get ANOVA table with Type III tests
anova_table <- anova(mixed_model, type = 3)
print(anova_table)
Type III Analysis of Variance Table with Satterthwaite's method
      Sum Sq Mean Sq NumDF DenDF F value  Pr(>F)  
TREAT   2434  811.33     3    12  2.7171 0.09126 .
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
# For comparison, also run a traditional nested ANOVA
nested_aov <- aov(ALGAE ~ TREAT + TREAT:PATCH, data = andrew)
std_summary <- summary(nested_aov)[[1]]

# Extract MS values - using exact row names
MS_treat <- std_summary["TREAT      ", "Mean Sq"] 
MS_patch <- std_summary["TREAT:PATCH", "Mean Sq"]
MS_residual <- std_summary["Residuals", "Mean Sq"]

# Print MS values to check
print("MS values:")
[1] "MS values:"
print(c(Treatment = MS_treat, Patches = MS_patch, Residual = MS_residual))
Treatment   Patches  Residual 
 4809.712  1770.162   298.600 
# Extract df values
df_treat <- std_summary["TREAT      ", "Df"]
df_patch <- std_summary["TREAT:PATCH", "Df"]
df_residual <- std_summary["Residuals", "Df"]


# Calculate correct F ratios for nested design
F_treat <- MS_treat / MS_patch
F_patch <- MS_patch / MS_residual

# Calculate p-values
p_treat <- pf(F_treat, df_treat, df_patch, lower.tail = FALSE)
p_patch <- pf(F_patch, df_patch, df_residual, lower.tail = FALSE)

# Create ANOVA table
trad_anova_table <- data.frame(
  Source = c("Treatment", "Patches (treatment)", "Residual"),
  df = c(df_treat, df_patch, df_residual),
  MS = c(MS_treat, MS_patch, MS_residual),
  F = c(F_treat, F_patch, NA),
  p = c(p_treat, p_patch, NA)
)

# Format p-values
trad_anova_table$p <- ifelse(trad_anova_table$p < 0.001, "<0.001",
                       ifelse(is.na(trad_anova_table$p), NA,
                              format(trad_anova_table$p, digits = 3)))
trad_anova_table
               Source df       MS        F           p
1           Treatment  3 4809.712 2.717102 0.091262004
2 Patches (treatment) 12 1770.162 5.928207      <0.001
3            Residual 64  298.600       NA        <NA>
# # Display traditional ANOVA table with flextable
# trad_anova_table %>%
#   flextable() %>%
#   set_header_labels(
#     Source = "Source of variation",
#     df = "df",
#     MS = "MS",
#     F = "F",
#     p = "p"
#   ) %>%
#   colformat_double(j = c("MS", "F"), digits = 2) %>%
#   autofit() %>%
#   add_header_lines("ANOVA table for nested design") %>%
#   theme_box()

Variance Components

We can extract the variance components from the mixed model:

# Print corrected results

# Extract variance components
vc <- VarCorr(mixed_model)
print(vc)
 Groups      Name        Std.Dev.
 TREAT:PATCH (Intercept) 17.156  
 Residual                17.280  
# Extract variance components
var_comp_patch <- as.numeric(vc$`TREAT:PATCH`)
var_comp_residual <- attr(vc, "sc")^2

# Calculate percentage of total variance
total_var <- var_comp_patch + var_comp_residual
pct_patch <- var_comp_patch / total_var * 100
pct_residual <- var_comp_residual / total_var * 100

# Calculate treatment variance component
n_quad <- 5  # Number of quadrats per patch
n_patch <- 4  # Number of patches per treatment
var_comp_treatment <- (MS_treat - MS_patch) / (n_quad * n_patch)

# Format variance components for display
var_comp_treatment_display <- ifelse(var_comp_treatment < 0, 
                                    paste0("(", format(abs(var_comp_treatment), digits = 2), ")"),
                                    format(var_comp_treatment, digits = 2))

# Create variance components table
var_comp_table <- data.frame(
  Source = c("Treatment", "Patches (treatment)", "Residual"),
  `Var.comp` = c(var_comp_treatment_display, 
               format(var_comp_patch, digits = 2),
               format(var_comp_residual, digits = 2))
)

# Display variance components table
var_comp_table %>%
  flextable() %>%
  set_header_labels(
    Source = "Source of variation",
    Var.comp = "Variance component"
  ) %>%
  autofit() %>%
  add_header_lines("Variance components") %>%
  theme_box()

Variance components

Source of variation

Variance component

Treatment

152

Patches (treatment)

294

Residual

299

# Complete table with all information
complete_table <- data.frame(
  Source = c("Treatment", "Patches (treatment)", "Residual"),
  df = c(df_treat, df_patch, df_residual),
  MS = c(MS_treat, MS_patch, MS_residual),
  F = c(F_treat, F_patch, NA),
  p = c(trad_anova_table$p[1], trad_anova_table$p[2], NA),
  `Var.comp` = c(var_comp_treatment_display, 
                format(var_comp_patch, digits = 2),
                format(var_comp_residual, digits = 2))
)

# Display complete table
complete_table %>%
  flextable() %>%
  set_header_labels(
    Source = "Source of variation",
    df = "df",
    MS = "MS",
    F = "F",
    p = "p",
    Var.comp = "Var. comp."
  ) %>%
  colformat_double(j = c("MS", "F"), digits = 2) %>%
  autofit() %>%
  add_header_lines("Complete ANOVA table with variance components") %>%
  theme_box()

Complete ANOVA table with variance components

Source of variation

df

MS

F

p

Var. comp.

Treatment

3

4,809.71

2.72

0.091262004

152

Patches (treatment)

12

1,770.16

5.93

<0.001

294

Residual

64

298.60

299

Interpretation of ANOVA Results

The nested ANOVA using mixed models reveals that there was no significant effect of urchin density treatment on algae cover (F = 2.72, df = 3, 12, p = 0.0913). However, there was significant variation among patches within treatments (F = 5.93, df = 12, 64, p < 0.001).

The variance component for patches nested within treatments (294) indicates substantial spatial heterogeneity in algae cover, highlighting the importance of accounting for this spatial variation in the analysis. The negative variance component for treatment suggests that there is more variation among patches within treatments than among treatments themselves.

Lecture 14: Post-hoc Comparisons

Although the main effect of treatment was not significant in the nested ANOVA (p = r format(p_treat, digits=3)), we can still examine the mean differences between treatments to understand patterns in the data. However, we should interpret these with caution given the lack of statistical significance at the α = 0.05 level.

# Calculate estimated marginal means
emm <- emmeans(mixed_model, ~ TREAT)

# Display EMMs with flextable
as.data.frame(summary(emm)) %>%
  flextable() %>%
  set_header_labels(
    TREAT = "Treatment",
    emmean = "Estimated Marginal Mean",
    SE = "Standard Error",
    df = "df",
    lower.CL = "Lower CL",
    upper.CL = "Upper CL"
  ) %>%
  colformat_double(j = c("emmean", "SE", "lower.CL", "upper.CL"), digits = 2) %>%
  autofit() %>%
  add_header_lines("Estimated marginal means for each treatment") %>%
  theme_box()

Estimated marginal means for each treatment

Treatment

Estimated Marginal Mean

Standard Error

df

Lower CL

Upper CL

Control

1.30

9.41

12

-19.20

21.80

66% Density

21.55

9.41

12

1.05

42.05

33% Density

19.00

9.41

12

-1.50

39.50

Removed

39.20

9.41

12

18.70

59.70

Lecture 14: Tukey Pairwise Comparisons

  • text
# Pairwise comparisons with Tukey adjustment
pairs <- pairs(emm, adjust = "tukey")
pairs
 contrast                  estimate   SE df t.ratio p.value
 Control - 66% Density       -20.25 13.3 12  -1.522  0.4553
 Control - 33% Density       -17.70 13.3 12  -1.330  0.5625
 Control - Removed           -37.90 13.3 12  -2.849  0.0615
 66% Density - 33% Density     2.55 13.3 12   0.192  0.9974
 66% Density - Removed       -17.65 13.3 12  -1.327  0.5646
 33% Density - Removed       -20.20 13.3 12  -1.518  0.4573

Degrees-of-freedom method: kenward-roger 
P value adjustment: tukey method for comparing a family of 4 estimates 
# # Display pairwise comparisons with flextable
# as.data.frame(summary(pairs)) %>%
#   flextable() %>%
#   set_header_labels(
#     contrast = "Contrast",
#     estimate = "Estimate",
#     SE = "Standard Error",
#     df = "df",
#     t.ratio = "t ratio",
#     p.value = "p-value"
#   ) %>%
#   colformat_double(j = c("estimate", "SE", "t.ratio", "p.value"), digits = 3) %>%
#   autofit() %>%
#   add_header_lines("Pairwise comparisons between treatments (Tukey-adjusted)") %>%
#   theme_box()

Lecture 14: Letter Display

# Extract compact letter display for plotting
cld <- multcomp::cld(emm, alpha = 0.05, Letters = letters)

cld
 TREAT       emmean   SE df lower.CL upper.CL .group
 Control        1.3 9.41 12   -19.20     21.8  a    
 33% Density   19.0 9.41 12    -1.50     39.5  a    
 66% Density   21.6 9.41 12     1.05     42.0  a    
 Removed       39.2 9.41 12    18.70     59.7  a    

Degrees-of-freedom method: kenward-roger 
Confidence level used: 0.95 
P value adjustment: tukey method for comparing a family of 4 estimates 
significance level used: alpha = 0.05 
NOTE: If two or more means share the same grouping symbol,
      then we cannot show them to be different.
      But we also did not show them to be the same. 
# # Display CLD with flextable
# as.data.frame(cld) %>%
#   flextable() %>%
#   set_header_labels(
#     TREAT = "Treatment",
#     emmean = "Estimated Marginal Mean",
#     SE = "Standard Error",
#     df = "df",
#     lower.CL = "Lower CL",
#     upper.CL = "Upper CL",
#     .group = "Group"
#   ) %>%
#   colformat_double(j = c("emmean", "SE", "lower.CL", "upper.CL"), digits = 2) %>%
#   autofit() %>%
#   add_header_lines("Compact letter display of treatment means") %>%
#   theme_box()

Interpretation of Treatment Comparisons The mean algae cover for the Control treatment (1.30%) appears considerably lower than for the reduced urchin density treatments (66% Density: 21.55%, 33% Density: 19.00%, Removed: 39.20%). While the visual pattern suggests an inverse relationship between urchin density and algae cover, with complete removal showing the highest algae cover, the nested ANOVA showed that these differences were not statistically significant at the α = 0.05 level (p = r format(p_treat, digits=3)). The high variability among patches within treatments likely contributed to the lack of statistical significance for the treatment effect.

Assumption Testing

For valid inference from mixed models, several assumptions must be met. We test these assumptions below.

Normality of Residuals

# QQ plot of residuals
qqnorm(resid(mixed_model))
qqline(resid(mixed_model))

# Histogram of residuals
hist(resid(mixed_model), main = "Histogram of Residuals",
     xlab = "Residuals", breaks = 15)

# More advanced residual diagnostics using DHARMa
sim_residuals <- simulateResiduals(fittedModel = mixed_model)
plot(sim_residuals)

Homogeneity of Variance

# Residuals vs. fitted values plot
plot(fitted(mixed_model), resid(mixed_model),
     xlab = "Fitted Values", ylab = "Residuals",
     main = "Residuals vs. Fitted Values")
abline(h = 0, lty = 2, col = "red")
# Levene's test for homogeneity of variance
levene_test <- leveneTest(ALGAE ~ TREAT, data = andrew)
levene_test
Levene's Test for Homogeneity of Variance (center = median)
      Df F value     Pr(>F)    
group  3  8.1694 0.00008785 ***
      76                       
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Interpretation of Assumption Tests

The Q-Q plot shows some deviation from normality, particularly in the tails, and Levene’s test indicates significant heterogeneity of variances across treatments (F = 8.17, p < 0.001). As noted in the original analysis, there were “large differences in within-cell variances” in this dataset, and transformations did not improve variance homogeneity.

The DHARMa residual diagnostics also indicate potential issues with the distribution of residuals and homogeneity of variance. The residuals vs. fitted plot shows a pattern of increasing variance with increasing fitted values, confirming the heteroscedasticity.

However, mixed models are generally robust to moderate violations of assumptions, especially with balanced designs. Since transformations were not effective in improving the data properties, analyzing the untransformed data is a reasonable approach in this case.

Post-hoc Comparisons

Although the main effect of treatment was not significant in the nested ANOVA (p = 0.0913), we can still examine the mean differences between treatments to understand patterns in the data.

# Calculate estimated marginal means
emm <- emmeans(mixed_model, ~ TREAT)
emm
 TREAT       emmean   SE df lower.CL upper.CL
 Control        1.3 9.41 12   -19.20     21.8
 66% Density   21.6 9.41 12     1.05     42.0
 33% Density   19.0 9.41 12    -1.50     39.5
 Removed       39.2 9.41 12    18.70     59.7

Degrees-of-freedom method: kenward-roger 
Confidence level used: 0.95 
# Pairwise comparisons with Tukey adjustment
pairs <- pairs(emm, adjust = "tukey")
pairs
 contrast                  estimate   SE df t.ratio p.value
 Control - 66% Density       -20.25 13.3 12  -1.522  0.4553
 Control - 33% Density       -17.70 13.3 12  -1.330  0.5625
 Control - Removed           -37.90 13.3 12  -2.849  0.0615
 66% Density - 33% Density     2.55 13.3 12   0.192  0.9974
 66% Density - Removed       -17.65 13.3 12  -1.327  0.5646
 33% Density - Removed       -20.20 13.3 12  -1.518  0.4573

Degrees-of-freedom method: kenward-roger 
P value adjustment: tukey method for comparing a family of 4 estimates 
# Compact letter display
cld <- multcomp::cld(emm, alpha = 0.05, Letters = letters)
cld
 TREAT       emmean   SE df lower.CL upper.CL .group
 Control        1.3 9.41 12   -19.20     21.8  a    
 33% Density   19.0 9.41 12    -1.50     39.5  a    
 66% Density   21.6 9.41 12     1.05     42.0  a    
 Removed       39.2 9.41 12    18.70     59.7  a    

Degrees-of-freedom method: kenward-roger 
Confidence level used: 0.95 
P value adjustment: tukey method for comparing a family of 4 estimates 
significance level used: alpha = 0.05 
NOTE: If two or more means share the same grouping symbol,
      then we cannot show them to be different.
      But we also did not show them to be the same.

Visualization

# Create boxplot with jittered points
ggplot_boxplot <- ggplot(andrew, aes(x = TREAT, y = ALGAE, fill = TREAT)) +
  geom_boxplot(alpha = 0.7, outlier.shape = NA) +
  geom_jitter(width = 0.2, alpha = 0.4, size = 1) +
  scale_fill_viridis_d(option = "D", end = 0.85) +
  labs(
    title = "Urchin Density effect on Algae Cover",
    x = "Urchin Density ",
    y = "Algae Cover (%)",
    caption = "Figure 1: Boxplots showing the distribution of algal cover across urchin density.\nDespite visual differences, the treatment effect was not statistically significant (p = 0.091)."
  ) +
  theme_minimal() +
  theme(
    legend.position = "none",
    plot.title = element_text(face = "bold", size = 14),
    axis.title = element_text(face = "bold", size = 12),
    axis.text = element_text(size = 10),
    plot.caption = element_text(hjust = 0, face = "italic", size = 10)
  )

# Create means plot with error bars
means_plot <- ggplot(summary_stats, aes(x = TREAT, y = mean, group = 1)) +
  geom_point(size = 3, shape = 21, fill = "white") +
  geom_errorbar(aes(ymin = mean - se, ymax = mean + se), width = 0.2) +
  labs(
    title = "Mean Algae Cover by Urchin Density",
    x = "Urchin Density",
    y = "Algae Cover (%)",
    caption = "Figure 2: Mean (± SE) percentage cover of algae across urchin density treatments."
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    axis.title = element_text(face = "bold", size = 12),
    axis.text = element_text(size = 10),
    plot.caption = element_text(hjust = 0, face = "italic", size = 10)
  )
# Display plots
ggplot_boxplot

means_plot

# Combined plot using patchwork
ggplot_boxplot + means_plot + plot_layout(ncol = 1)

Discussion

Scientific Interpretation

Our mixed model analysis of the nested design revealed substantial spatial heterogeneity in algae cover, with significant variation among patches within each treatment (p < 0.001). Surprisingly, the effect of urchin density treatments on filamentous algae cover was not statistically significant at the α = 0.05 level (p = 0.091), despite apparent trends in the data.

The descriptive statistics show a pattern where algae cover appears to increase as urchin density decreases, with the Control treatment (mean = 1.3%) showing minimal algae cover compared to reduced density treatments (66% Density: 21.55%, 33% Density: 19.00%, and Removed: 39.20%). This pattern suggests a potential density-dependent relationship between urchin grazing and algal abundance, but the high variability among patches masked the treatment effect.

The substantial variance component associated with patches nested within treatments (294.31, approximately 39.5% of total variance) underscores the importance of spatial heterogeneity in structuring algal communities. This finding highlights the necessity of accounting for spatial variability when designing and analyzing ecological field experiments.

From an ecological perspective, these results suggest that while sea urchins may influence algal communities through grazing, local environmental factors and patch-specific conditions play a dominant role in determining algae cover. This has important implications for ecosystem management, as it indicates that the effects of urchin density manipulations may be context-dependent and influenced by local environmental conditions.

Comparison with Traditional Nested ANOVA

The linear mixed model approach provides similar results to the traditional nested ANOVA approach. The main advantage of the mixed model is the more elegant handling of random effects and the extensive diagnostic tools available through packages like DHARMa.

The mixed model approach confirms that:

  1. Treatment effects are not significant (p = 0.091)
  2. Patches within treatments show significant variation (p < 0.001)
  3. The variance components are similar to those from the traditional approach

In both methods, the key ecological finding is the strong spatial heterogeneity in algal cover that overrides the grazing effect of urchins at different densities.

References

Andrew, N. L., & Underwood, A. J. (1993). Density-dependent foraging in the sea urchin Centrostephanus rodgersii on shallow subtidal reefs in New South Wales, Australia. Marine Ecology Progress Series, 99, 89-98.

Quinn, G. P., & Keough, M. J. (2002). Experimental design and data analysis for biologists. Cambridge University Press.